[SERVER-31880] Re-run TCMalloc analysis with YCSB workload Created: 08/Nov/17  Updated: 30/Oct/23  Resolved: 11/Dec/17

Status: Closed
Project: Core Server
Component/s: None
Affects Version/s: None
Fix Version/s: 3.7.1

Type: Task Priority: Major - P3
Reporter: Gregory McKeon (Inactive) Assignee: Mark Benvenuto
Resolution: Fixed Votes: 0
Labels: None
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Backwards Compatibility: Fully Compatible
Sprint: Platforms 2017-12-04, Platforms 2017-12-18
Participants:

 Comments   
Comment by Mark Benvenuto [ 11/Dec/17 ]

The Yahoo Cloud Serving Benchmark (YCSB) is an alternate benchmark that treats MongoDB as a key value store. It has two phases: a load phase, and a run phase. During the load phase a configurable number of documents are loaded into MongoDB. During the run phases, a mix of updates, and document reads are performed for a fixed number of operations.

The source for the variation used is here: https://github.com/mongodb-labs/YCSB. Two variations were used: 95% Read/ 5% Update, and 50% Read/ 50% Write. 2000000 records were used, and 200,000,000 operations except for the poorly performing tests where it was limited to 20,000,000 (so the test would finish in minutes instead of hours). I ran three variations 48, 256, and 512 threads.

Overall, YCSB follows a similar pattern to the John Page workload against 3.2.16. The most significant difference between the two tests is that YCSB does not decay over time like the John Page workload. This is because YCSB simply does not insert records during its run phase.

Generated at Thu Feb 08 04:28:28 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.